Fix lost 0. index prefix for ListSerializer errors on DRF 3.18

[sshishov]

Summary

Fixes #121.

With DRF 3.18, ListSerializer.to_internal_value collects child errors in a dict keyed by the integer index of the item (e.g. {0: {"email": [...]}}) instead of a positional list as before. In flatten_errors, the dict branch only prefixed a child key with the parent attr when if attr: was truthy. Since the integer 0 is falsy in Python, an error on the first item (index 0) of a many=True payload lost its 0. prefix, while items at index 1 and above were prefixed correctly (any other int is truthy).

• Before: ['email'] for an error on item 0, ['1.email'] for item 1
• After: ['0.email'] for item 0, ['1.email'] for item 1

Fix

Check attr is not None instead of relying on attr's truthiness, so an attr of 0 is still treated as a valid prefix segment.

Test plan

• Added test_list_serializer_errors_indexed_dict in tests/test_flatten_errors.py, covering the DRF 3.18-style index-keyed error dict.
• Ran the full test suite locally with djangorestframework==3.18.0 — all 124 tests pass.
• Verified against the exact reproduction script from the issue; output now matches the expected ['0.email'] / ['1.email'].

🤖 Generated with Claude Code